As captured by Moore's Law, there is a longstanding trend in the semiconductor industry toward higher device densities and correspondingly smaller device geometries. The ellipsometry and reflectometry optical metrology tools that are used to evaluate semiconductor wafers must be able to respond to these ever decreasing device geometries by making measurements within smaller and smaller areas. Accordingly, a problematic circumstance has now arisen such that the size of the desired measurement area or measurement "pad" is comparable to the size of the optical spot used to make the measurement. An example of this circumstance is illustrated in FIG. 1. When this circumstance occurs, the measurement of the pad can be influenced by light reflected from the surrounding material. This influence in turn corrupts the accuracy and repeatability of the measurement because the surrounding material has different optical properties and a different depth than does the measurement pad itself.
In the case of the FIG. 1 example, the desired measurement would be of the thin layer 10 at the bottom of the well. The thin film layer 10 might be composed of a thin gate oxide layer, for instance. Crowding the beam is a top layer 20 that could well be composed of an entirely different material with significantly different optical properties than the layer to be evaluated.
In order to effectively apply optical metrology to semiconductor wafers, it is extremely important to be able make measurements with great precision and accuracy. For example, it is not uncommon to require layer thickness measurements to be repeatable to less than 0.1 .ANG. (1/30.sup.th of an oxide monolayer). Such very high precision requires that the nature of the reflected light be known with corresponding precision.
When the spot size of the measurement beam approaches the size of the measurement pad, accurate positioning of the sample becomes especially critical. Typical sample stages used in metrology tools have positioning uncertainties on the order of a few microns, largely as a result of backlash in the mechanical linkages of the stage or drive mechanism. A few microns can constitute a significant percentage of the dimensions of the measurement pad. Thus, in order to reliably localize the optical spot of the measurement beam entirely within the pad to be evaluated, one needs to either reduce the size of the optical spot, or to take steps to somehow get around the stage inaccuracy. Although pattern recognition systems (in which the sample image is compared to a stored image of the target area) can reduce the positioning uncertainty somewhat, the physical limitations of the stage hardware are always present at some level.
It also bears noting that even if the optical spot appears to be small in terms of the usual definitions of spot size (e.g., the 1/e.sup.2 beam width or Gaussian radius), there are often very faint tails that can extend well beyond these definitions. Given the extreme sensitivities required to accurately measure the thicknesses of very thin films, these very faint effects can cause an unacceptably large error. For example, we have found that for a Gaussian beam (typical of a well-focused laser) faint tails extending out to 3 times the Gaussian radius will corrupt the signals to an unacceptable degree. While it may be possible to "flatten" the beam profile to minimize the presence of such faint Gaussian tails using an aperture or other diffractive element, such techniques would tend to create undesirable interference fringes along the optical path within the instrument.
In fact, although shrinking the optical measurement spot size may generally be desirable in itself, such shrinking always comes at some cost. If the measurement beam source has a broad spatial extent (such as a tungsten filament or the arc of an arc lamp), then the light intensity at the sample surface tends to have an upper limit such that shrinking the optical spot means lowering the total amount of light. In turn, lowering the total light available for measuring tends to degrade the performance characteristics of the instrument because of decreased signal to noise ratios. Even if the measurement beam is bright and well collimated (such as a laser beam), the optical spot size will be still be limited by the power and complexity of the focusing lenses used to focus the beam on the sample surface. For a given beam diameter, shrinking the spot size means decreasing the focal length of the focusing lenses. This means crowding the lenses closer to the sample which, for off-axis optical systems, is a major inconvenience. This is so because typically one eventually runs into either the sample or some other optic used in the tool (e.g., a normal incidence lens used for a pattern recognition system). In addition, these high-numerical aperture lenses tend to be more prone to aberrations, and the larger curvatures can adversely impact the sensitive optical phase measurements needed for ellipsometry.
Once the practical limit for the optical spot size is reached, the only ways to minimize the effects of the surrounding material are either to somehow reduce the uncertainties in stage positioning or else to take into account the surrounding material in the analysis of the measured data. One aspect of the present invention is directed to a means of minimizing the stage positioning uncertainties by using a novel technique for finding the center of the measurement pad. This technique takes advantage of the fact that while the absolute accuracy of a positioning stage may be as poor as several microns, the ability of the stage to make incremental movements is much finer. In another aspect, the present invention is directed to a novel method for taking into account the effects of the surrounding material in analyzing the measured data.